A MODULAR APPROACH TO THE KORBA AQUIFER SEAWATER INTRUSION STUDY, 1, GIS FIELD DATA ANALYSIS
A. Giacomelli and C. Paniconi
Environment Group
Centro di Ricerca, Sviluppo e Studi Superiori in Sardegna (CRS4) Cagliari, Italy
I. Khlaifi and J. Tarhouni
Département de Génie Rural, Eaux et Forêts Institut National Agronomique de Tunisie (INAT)
Tunis, Tunisia ABSTRACT
Seawater intrusion is an important environmental problem in the coastal aquifers of many Mediterranean countries. In the 438 km2 Korba aquifer in eastern Tunisia, a large increase in the number of pumping wells for irrigation purposes since the 1960s has resulted in a lowering to below sea level of the water ta- ble in several observation piezometers, and in a consequent deterioration of the water quality. Several re- mediation scenarios are being considered for this region, including rationalization and control of water pumping from the wells, artificial recharge of the aquifer and construction of small dams to serve as an alternative source of irrigation water. In order to investigate the impact of these measures on the aquifer water quality, a GIS-based modeling study is being undertaken. The available data set comprises a number of layers of geographical information, giving a complete hydrogeological characterization of the region, and time series of chemical and hydrologic variables acquired during several ground sampling campaigns performed in the last thirty years. The GIS is used to organize this heterogeneous data structure and to control the data flow through various phases of the work, i.e. the pre-processing of input data for the mod- el, the interpretation of model outputs, and the calibration of the model itself. The GIS serves also as a support tool in the generation of the 3-D computational grid used in the numerical simulations. In addition to describing the data structure and the organization of the system, the paper illustrates also the implemen- tation of a simple recharge optimization scenario.
INTRODUCTION
An accurate description of the spatial and tem- poral dynamics of seawater intrusion is of criti- cal importance in the study of aquifers affected by this type of contamination. A distributed flow and transport model, supported by an adequate knowledge of aquifer characteristics, can be a
valuable tool to corroborate and extrapolate the observations provided by monitoring networks.
Most importantly, distributed models allow a detailed simulation of the evolution of the intru- sion process under different conditions, and – thus– the evaluation of the effect of different proposed remediation strategies. However, the vast amount and the heterogeneity of the data required for a study on a real site poses on the
modeler a heavy burden, in terms of data pro- cessing and management, which is added to the actual modeling issues and which may greatly limit the modeling results.
Geographic information systems (GIS) have by now been well recognized as an ideal and almost indispensable support to environmental model- ing (Hornberger and Boyer, 1995, Nyerges, 1993, Maidment, 1996). This is due to their ca- pacity of handling georeferenced information and of offering the possibility of performing complex spatial analysis and visualization tasks within a consistent framework. Integrated appli- cations of GIS and models may span from sys- tems specifically dedicated to support the simu- lation process (Harris, 1993, Nachtnebel 1993), to complete implementations of decision support systems (Fuerst et al., 1993) for water resources management. However, GIS do have some limi- tations which require still additional tools for a comprehensive evaluation of the results of a complex groundwater model, such as the one adopted in this study. These tools typically in- clude scientific visualization packages, for the intuitive representation of multi-dimensional
data sets, and database management systems.
The approach followed in the integration of such diverse tools represents a study topic per se, since the effort required by the integration pro- cess is significant and must be addressed by ap- propriate strategies (Paniconi et al., 1998).
In this article, having given a description of the Korba site and of the data available for the study, we proceed to show how GIS has been used in the organization of the geographic data set. The analytic and graphic functionality of GIS, to- gether with the customization capabilities of- fered by the available scripting languages have been exploited to develop a set of routines for the exchange of data with the numerical model, greatly simplifying and enhancing common tasks in the modeling process. These include the gen- eration of the computational mesh, the definition of the boundary and initial conditions, and the visualization of simulation results. Finally, the possibility of exploiting the GIS/model frame- work for the analysis of remediation scenarios (which is the ultimate objective of this study) is briefly discussed.
Figure 1. The Korba aquifer: location and simplified geologic map, indicating also the river network and the position of the climatic stations.
DESCRIPTION OF THE STUDY SITE
The Korba aquifer is a part of the western coastal aquifer of the Cap Bon area, extending in the plain from the city of Ras Maamoura in the south to the city of Kelibia in the North, and is bounded by the Mediterranean Sea in the east and the Djbel Sidi AbedErrahmen anticline in the west. The aquifer covers an area of 438 km2 and was formed during the Pliocene and Quater- nary ages by sedimentation of eroded products from the Djbel Sidi AbedErrahmen anticline.The water table is located in the Plio-Quaternary formations and the aquifer substratum is consti- tuted by underlying Miocene marl formations.
Two distinct formations compose the Korba aq- uifer: a Pliocene sandstone whose stratigraphic series correspond to an alternation of sandstone and marl, and a Quaternary alluvium containing detrital sediment (sand, gravel, silt) with thin clay lenses. The sandstone formation spans the entire aquifer and has a mean depth of 85 m, while the detrital formation occurs in the south- western part of the aquifer and has a thickness that varies between 20 and 25 m. Aquifer depth is highly variable, ranging from 150 m in the south to 30 m in the north, and decreasing from east to west to nearly vanish at the Djbel Sidi AbedErrahmen anticline. A characteristic geo- morphologic element in the area is the presence of a dune formation of quaternary sediments running parallel to the coastline and separating the Korba plain from the sea, characterized by a high hydraulic transmissivity.
The main hydrographic network consists of three rivers with irregular flow (Owed), which are completely dry in the summer. Precipitation on the Korba plain is monitored by five stations displaced in the area, and exhibits a very irregu- lar and temporal distribution. According to the Korba station data, 60% of the annual rainfall is concentrated between November and March, and the rainfall deficit covers a period of nearly 10 months, reaching its maximum (150 mm) in July and August. Mean annual rainfall computed over an observation period of 28 years (1964 to 1996), is around 460 mm, and in exceptionally dry years rainfall may drop to one third of this value. The location of the Korba site and a repre-
sentation of the described features are shown in Figure 1.
Natural recharge of the aquifer is insured by rainfall infiltration, for which an average of 32 mm/year has been estimated on the plain, with higher values in the river beds and in the dune areas, and lower values in the Quaternary areas.
The Plio-quaternary aquifer receives its main natural recharge from the piedmont of Djebel Abderrahmane.
The Korba aquifer is exploited by 7309 wells extracting in all nearly 47106 m3 of water per year, against a permitted exploitation in the range of 50106 m3/year (Ennabli, 1977). The distribution of the wells is particularly dense to- wards the coastal part of the plain, determining a distribution of water exploitation which tends to accentuate the movement of the freshwa- ter/saltwater interface.
Monitoring of the Korba aquifer was begun in 1962. The over-exploitation of the aquifer is re- flected in a steady lowering of the piezometric level observed in the monitoring network (from 0 to –5 m) and in a consequent decrease in the yield of the pumping wells. At the same time, a continuous growth of water salinity has been observed in the eastern part of the site, with peak concentration values of 5 to 8 g/l.
AVAILABLE DATA
The data available for the study can be divided in three sets, respectively giving the hydrogeo- logic characterization of the site, the information on the exploitation of the aquifer, and the moni- toring of the seawater intrusion effects.
Cartographic data are available for geology, hy- draulic transmissivity, hydrography, topography and pedology. Together with a map of the aqui- fer limits obtained from a previous study (Enna- bli, 1977) and information on the geometry of the base of the aquifer, these data are used to define the computational domain for the model and to provide a description of the physical properties of the aquifer.
The data on the exploitation of the aquifer is rep- resented by a map indicating the position of clus- ters of wells (regularly distributed on the plain), to which values of exploitation are linked, repre-
senting the sum of the exploitation of the real wells. In addition, a map of the public irrigated zones is available, from which an indication on the eventual infiltration due to irrigation may be derived.
he third set of data concerns monitoring infor- mation. This is represented by a map giving the position of 99 monitoring wells located on the site. The monitoring well map is linked to rec- ords of piezometric level and various chemical parameters. Piezometry and salinity contour line maps derived from the monitoring data prior to this study are available, covering a significant part of the plain, for 1962, 1970, and 1996.
USE OF GIS IN THIS STUDY
Two main GIS-based tasks can be identified in this study: an initial editing phase, for the crea- tion of the geographic data set from the available raw data, and an analysis phase, with a strong interaction between GIS and numerical model.
The ARC/INFO package has been used to digit- ize the geographic features from the source map sheets and to assign attribute information to the data layers created, while the interaction with the model takes advantage also of the graphical user interface provided by the ArcView package, and on a set of in-house developed FORTRAN 90 routines. The resulting data framework, illustrat- ed in Figure 2, shows how the GIS and the mod-
el act as two loosely integrated units, connected through a layer represented by the pre- and post- processing modules.
Creation of the geographic data set
All geographic data used in this study was ini- tially available in paper format, in different scales and deriving from different sources. The first step was to digitize from the cartographic media the required information, and –for each of the required themes– to join the data deriving from separate map sheets into layers covering the whole region. In this process several con- sistency issues had to be faced and solved (e.g.
differing classifications for geologic map sheets dating back to different years). In general, the aim was to create a faithful replication of the original data set, from the point of view both of geometry and of attribute data.
The salinity and piezometry maps, as well as the transmissivity map –which were taken from pre- vious studies on the site– did not contain any geographic coordinate reference, and thus need- ed to be georeferenced on the more reliable data layers, taking the control points from the coast- line, which was the only recognizeable geo- graphic feature. This map, in conjunction with the geologic map, was used to define the surface limits of the aquifer.
Figure 2. A diagram representing the flow of data through the components of the sys- tem.
A 50x50 m digital elevation model for the study was obtained by interpolation of the topographic data.
2D mesh generation
The finite element mesh used by the CODESA- 3D model (presented in part 2 of this paper) is derived from a 2D mesh generated using the module available in ARC/INFO for the creation of triangulated irregular networks, or TINs (En- vironmental Systems Research Institute, 1988).
The advantage of this approach, compared to the use of other non GIS-based mesh generators, is that a direct link is maintained between the mesh structure and the (geographic data) input on which it is based. As a consequence, the user can benefit from a more intuitive identification of the important features, which may either be repre- sented by the “primitive” data layers, or by the result of complex spatial analyses, combining different information.
The TIN data structure holds information on ge- ometry (coordinates of all nodes, and connectivi- ty, defining arcs and triangles), and on a parame- ter representing the z dimension (typically eleva- tion), which is automatically interpolated at all
nodes, given a set of points with assigned z and a description of existing discontinuities and “clip”
areas (which can be used to exclude parts of the domain from the interpolation), thus allowing to incorporate the most diverse geometric features and to reproduce even very complex surfaces.
All the information held in a TIN can be export- ed to a simple ASCII file format, which can easi- ly be processed to obtain the input data for the model.
The main problem which has been encountered in this approach is that the TIN module provides only basic means to assess the quality of the re- sulting mesh with respect to regularity and asso- ciated numerical problems. In the case of an Eu- lerian model, and in order to eliminate numerical dispersion and instability, the mesh must satisfy the Péclet criterium (Huyakorn, 1984). Further- more, an indirect influence on the conditioning may be given by the presence of patch elements, i.e. clusters of many elements connected at one node (Kunianski, 1993).
Since the GIS-supported mesh generation re- quires a fraction of the time compared to manual design or use of structural mesh generation packages, the strategy adopted in this study was
U
U
U U U U U U
U
U
U
U U U U U U U
U
U U U U
U U
U U U
U
U
U
U
U U U U
U U
U
U U U
S S S
S S S
S S S S S
S S S S S
S S S S S S S S
S S S S S S S S S
SS S S S S S S S S S S
SS S S S S S S S S S
S S S S S S S S S S S S S S
S S S S S S S S S S S S
S S S S S S S S S S S S S S
S S S S S S S S SS S S
S S S S S S S S S S
SS S SS S S S S
S
S S S S S S S S S S
S S S S S S S S
SS S S S S S S S S S S S
SS S S S S S S S S
S S S S S S S S S S S S S S
S S S S S S S S S S S S
SS S S S S S S S S S S S S
SS S S S S S S S S S S S
SS SS S S S S S S S
S S S S S S S S S S
S S S S S S S
S S S S S S
S S S S
S
S S
S
S
S S S S S S
S
S
S
S S S S S S
S
S
S S S S
S S
S S S
S
S
S
S
S S S S
S S
S
S S S
% % % % % %
% % % % % %
% % % % % % %
% % % % % % %
% % % % % % %
% % % %
% % % %
% % % %
% % % %
% % % %
% % %
U Monitoring wells Model4_p ol
% Elevation grid Limit of the domain
N
500 0 500 Meters
Figure 3. A detail of the 2D mesh created using ARC/INFO. The mesh configuration and resolution was in part determined bt the location of pumping and monitoring wells, which were made to coincide with grid points.
to base the selection of the 2D mesh on an itera- tive process, with a first qualitative assessment of the TIN output, directly followed by simula- tion tests to highlight numerical problems, which were in turn used to modify the input for the mesh generation.
Even though the TIN module has a number of control parameters, which can be used to specify how nodes are to be extracted from each of the input data layers (and how the z values are to be interpolated), it has been observed that in the case of the elevation contour lines it was not possible to obtain a satisfactory distribution of nodes. It was then decided to derive the nodes from the raster DEM, imposing a sampling dis- tance of 500 m. In addition to elevation, the fea- tures considered in the definition of the 2D mesh are: the pumping and monitoring wells, the lim- its of the computational domain (based on geo- logic outcrops), and a reclassified transmissivity map.
Once the 2D mesh is generated, the layer hold- ing the aquifer depth information is automatical- ly queried in correspondence of the element nodes, outputting the aquifer depth values neces- sary to define the 3D computational domain. The user is guided in this process by means of a rou- tine developed in AML, the scripting language used by ARC/INFO, while the conversion from the TIN ASCII export format to the format re- quired by the model is performed via a set of FORTRAN 90 routines.
Definition of boundary and initial condi- tions
In addition to defining the geometry of the 2D mesh, GIS functionality has been used to define the boundary and initial conditions for the simu- lations. Once the TIN has been created, its ele- ments can be converted to geographic data cov- erages (either in point format, representing the nodes, or in polygon format, representing the triangles). The identifiers of the elements in these coverages are consistent with the identifi- ers of the TIN and of the 2D mesh. This allows to transfer physical parameters (or the result of any spatial processing on the available coverag- es) to the mesh. In this task the user takes full
advantage of the graphical user interface provid- ed by Arcview.
Support to analysis of simulation results
The output of a simulation is fed to a post- processing module composed of a second set of FORTRAN 90 routines and AMLs, converting the numerical results back to ARC/INFO TIN and coverage formats, which can be used to compare the results of different simulation runs and the data acquired from the monitoring net- work. In particular, piezometry contour lines are derived from the TINs for an immediate compar- ison with field data. Advanced three-dimensional visualization of the results is performed outside the GIS environment, employing a scientific vis- ualization package, as presented in part 2 of this paper.Analysis of remediation scenarios
Several remediation scenarios are being consid- ered for the Korba region. These include ration- alization and control of water pumping from the wells, artificial recharge of the aquifer, and con- struction of small dams to serve as an alternative source of irrigation water. The evaluation of these scenarios requires a combined use of the spatial analysis capabilities offered by the GIS and of the simulation model.
The criteria defining optimal remediation strat- gies can be applied to the site by means of GIS- based spatial analysis functions and algorithms.
For instance, it may be possible to define the position of one or more reservoirs, minimizing the cost of water distribution and maximizing the extent of the areas which can be subject to re- charge with their supply. The scenario thus iden- tified then needs to be translated into an appro- priate set of boundary conditions for the numeri- cal model and –finally– results from the simula- tion can be brought back to the GIS, to be ana- lyzed and compared with other scenarios.
In the framework of this study it has not been possible to undertake an exhaustive analysis of different scenarios or to implement sophisticated multi-criteria methodologies, and only a prelimi- nary test has been performed. This takes into account the geomorphology of the site, and iden-
tifies areas characterized by a higher suitability for artificial recharge, estimated as a function of hydraulic transmissivity, geology, and local slope. To this end, the corresponding geographic data layers have been reclassified, passing from the value of the initial parameter to a qualitative index representing the suitability; the three indi- ces have then been summed, leading to the final estimate (Figure 4). The pumping wells corre- sponding to the high suitability areas can be au- tomatically identified and the appropriate boundary conditions for the model are assigned.
CONCLUSIONS
The possibility of supporting a distributed model with a robust geographic information system, consisting of a well-designed geographic infor- mation data set and of software facilities that provide a seamless data exchange between the GIS, the model, and the visualization packages, greatly enhances any environmental modeling study. The most substantial benefits of a GIS support to the model may be summarized in the
possibility of manipulating data in an extremely direct and intuitive manner (i.e. the modeler can think in terms of physical features), leaving to the data infrastructure the burden of transform- ing the results of his spatial analyses in the form required by the model. In addition, heavy tasks – such as mesh generation or assignment of com- plex boundary conditions– can be automated and performed in a fraction of the time usually need- ed in absence of a GIS support. On the other hand, even though the trend is towards a grow- ing integration between different GIS and mod- els, the effort required to set up a an integrated framework remains very consistent.
The production of the geographic data set for the Korba aquifer, in addition to providing the data source for the numerical model, represents per se an important outcome of this study, since it has allowed the merging of a vast amount of infor- mation, collected from diverse sources and over many years and thus characterized by inconsist- encies which could have been untreatable with- Figure 4. Result from a preliminary recharge analysis, undertaken for the central part of the Korba site. A lighter tone indicates a higher suitability for artificial recharge. The dots represent the pumping wells located in areas of high suitability. The outermost lines indicate the limits of the computational domain.
out an adequate GIS platform for data manage- ment.
The analysis of alternative remediation scenarios for the Korba site needs to be carefully ad- dressed, and only preliminary tests have been undertaken in this study. In addition to a more comprehensive data set including information on data uncertainties, gaps, and errors, it should be noted that a robust remediation analysis also re- quires a properly calibrated model.
ACKNOWLEDGEMENTS
This work has been supported by the European Commission through grant number AVI-CT93- 2-073 and by the Sardinia Regional Authorities.
BIBLIOGRAPHY
Ennabli, M. (1977). Etude sur modèle mathé- matique des aquifères du Nord-Est de la Tunisie, PhD thesis, CIG-Ecoles des Mines, Paris, France.
Environmental Systems Research Institute (1992). Surface Modeling with TIN, ARC/INFO user guide. Redlands, California.
Fuerst, J., G. Girstmair, and H.P. Nachtnebel (1993). Application of GIS in decision support systems for groundwater management. In pro- ceedings of HydroGIS 93: Applications of Geo- graphic Information Systems in Hydrology and Water Resources. IAHS Publ. no. 211, 13-21.
Harris, J., S. Gupta, G. Woodside, and N. Ziem- ba (1993). Integrated use of a GIS and a three- dimensional finite-element model: San Gabriel basin groundwater flow analysis. In Environ- mental Modeling with GIS. Edited by M.F.
Goodchild, B.O. Parks and L.T. Steyaert, 168—
172, Oxford University Press.
Hornberger, G.M., and E. W. Boyer (1995). Re- cent advances in watershed modeling. In U.S.
National Report to International Union of Geod- esy and Geophysics 1991-1994: Contributions in Hydrology, edited by R.A. Pielke Sr. and R.M.
Vogel, 949-957, American Geophysical Union, Washington, DC.
Huyakorn, P.S., D.S. Thomas, and B.J. Thomp- son (1984). Techniques for making finite ele- ments competitive in modeling flow in variably
saturated porous media. Water Resour. Res., 20, 1099-1115.
Kuniansky, E.L., and R.A. Lowther (1993). Fi- nite-element mesh generation from mappable features. Int. J. Geographical Information Sys- tems, 7 (5), 395-405.
Maidment, D.R. (1996). GIS and hydrologic modeling – an assessment of Progress. Presented at the Third International Conference on GIS and Environmental Modeling, Santa Fe, New Mexico.
Nachtnebel, H.P., J. Fuerst, and H. Holzmann (1993). Application of geographical information systems to support groundwater modelling. In proceedings of HydroGIS 93: Applications of Geographic Information Systems in Hydrology and Water Resources. IAHS Publ. no. 211, 653- 664.
Nyerges, T.L. (1994). Understanding the scope of GIS: its relationship to environmental model- ing. In Innovations in GIS, 75-93, edited by M.F.
Worboys. Taylor & Francis.
Paniconi, C., S. Kleinfeldt, J. Deckmyn, and A.
Giacomelli (1998). Integrating GIS and data vis- ualization tools for distributed hydrologic mod- eling. Transactions in GIS (to appear).
